Silicon Photonic Neural Networks

Abstract

Microelectronic computers have encountered challenges in meeting all of today's demands for information processing. Meeting these demands will require the development of unconventional computers employing alternative processing models and new device physics. Neural network models have come to dominate modern machine learning algorithms, and specialized electronic hardware has been developed to implement them more efficiently. A silicon photonic integration industry promises to bring manufacturing ecosystems normally reserved for microelectronics to photonics. Photonic devices have already found simple analog signal processing niches where electronics cannot provide sufficient bandwidth and reconfigurability. In order to solve more complex information processing problems, they will have to adopt a processing model that generalizes and scales.

Neuromorphic photonics aims to map physical models of optoelectronic systems to abstract models of neural networks. It represents a new opportunity for machine information processing on sub-nanosecond timescales, with application to intelligent RF signal processing, mathematical programming, and real-time control. The strategy of neuromorphic engineering is to externalize the risk of developing computational theory alongside hardware. The strategy of remaining compatible with silicon photonics externalizes the risk of platform development. We demonstrate small neuromorphic photonic systems after developing the requisite new protocols, methods, and strategies for experimental proof-of-concept. The microring weight bank is introduced as the novel device that configures connection strengths between neurons.

The primary result of this thesis is a move from neuron-like photonic devices to complete networks of photonic neurons. This dissertation focuses on one kind of neuromorphic photonic network that is fully compatible with contemporary silicon photonic foundries. We give sufficient background on silicon photonics and neural networks at a level intended to introduce researchers from one field to the other. We describe design principles of silicon photonic neural networks and then derive scalability limits and power scaling relationships that connect performance to platform. Example benchmark tasks that demonstrate key processing capabilities are studied. Finally, we argue that the recurrent silicon photonic neural network is a good baseline against which to compare other implementations of neuromorphic photonics.